Image Annotation through Adaptive Dependency Fusion

摘要

In order to improve the performance of image annotation, recently proposed methods build their model combining multiple dependencies from relations between image and label (image/label), between images (image/image) and between labels (label/label). However, most of these methods cannot make multiple dependencies work jointly, and their performances is largely depending on the results predicted by image/label dependency. To address this problem, we propose an end-to-end image annotation model to associate these dependencies with the prediction path, which is composed of a series of labels in the order they are detected. Specially, our model can adaptively adjust the prediction path: from those easy-to-detect relevant labels to these hard-to-detect relevant ones. To validate the effective of the model, we conduct experiments on three well-known public datasets, COCO 2014, IAPR TC-12 and NUSWIDE, and achieve better performance than the state-of-the-art methods.